fix(qa): #1040 — Angry-DM scorer 'hang' was a too-short timeout (lens is SLOW, ~400s)

[100yenadmin]

Root cause (confirmed, not guessed)

A detached single claude -p probe (csmed-1 angrydm, no concurrency, 580s timeout) completed: EXIT=0, duration_ms=401324 (~402s), num_turns=1 (single-turn generation — no tool loop), valid 7.7 KB scorecard.

So the "#1040 combat-scorer hang" is the Angry-DM 5e-fidelity lens being SLOW, not stuck: rubric_angry_dm.md is ~32 KB (~3× tolkien, ~8× the mechanical rubric), and grading a combat-dense transcript against every 5e rule legitimately takes ~400s. The default 300s score.sh timeout killed it mid-generation → empty stdout that looked like a hang.

Concurrency made it worse, not caused it: run_duo.sh scored all 3 lenses with &, so the heavy angrydm shared API throughput with the 2 light lenses and ran even slower → exceeded 300s every combat run (the opus RRI lost its mech lens exactly this way: angrydm overall= blank while tolkien=4.0 scored). Social transcripts score in <300s, which is why only combat-dense runs "hung."

This corrects my earlier concurrency hypothesis — the clean (no-concurrency) probe still "hung" at a 240s timeout, which refuted concurrency-as-cause and pointed straight at slow-vs-timeout.

Fix (both halves)

• score.sh: default WORLDOS_SCORE_TIMEOUT 300 → 600 (covers ~400s + headroom). Fast lenses finish in ~60–150s, so the looser bound only matters for the heavy lens or a genuinely stuck call. Also updated the error-message default + the rationale comment.
• run_duo.sh: score the 2 light lenses (mechanical, tolkien) in parallel, wait, then score Angry-DM alone → full API throughput, lands near the ~400s baseline.

Validation

• bash -n both + score.sh guard-only mode (offline) OK.
• The 402s probe proves the lens completes given the timeout. An end-to-end score.sh re-score of csmed-1 angrydm with the new 600s default is running to confirm the wrapper path.

Impact

Unblocks clean mechanical measurement on combat-heavy runs (was n/a on the opus RRI duo). No engine/product change; QA-infra only.

Summary by CodeRabbit

• Chores
  • Improved scoring system reliability by optimizing concurrent lens execution and increasing timeout thresholds to prevent evaluation failures.
  • Enhanced diagnostic reporting to accurately reflect actual timeout durations in scoring logs.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 875661dc-e31b-41d8-9718-c2585ddcd936

📥 Commits

Reviewing files that changed from the base of the PR and between 427b366 and 865eb02.

📒 Files selected for processing (2)

• qa/run_duo.sh
• qa/score.sh

---

📝 Walkthrough

Walkthrough

The scoring phase in qa/run_duo.sh is changed so the mechanical and Tolkien lenses run concurrently and are awaited before the Angry-DM lens runs alone. In qa/score.sh, the claude -p timeout default is raised from 300s to 600s, and the empty-stdout retry diagnostic is updated to reference the new default.

Changes

Angry-DM Lens Isolation and Timeout Fix

Layer / File(s)	Summary
Lens execution sequencing qa/run_duo.sh	Mechanical and Tolkien score scripts are backgrounded together and awaited; Angry-DM score script then runs alone, replacing the prior single concurrent launch of all three lenses. Comments are updated to explain the sequencing rationale.
Scorer timeout and empty-output diagnostic qa/score.sh	Default claude -p timeout raised from 300s to 600s via ${WORLDOS_SCORE_TIMEOUT:-600}; inline comments document expected lens runtimes; the empty-stdout retry message now reports the updated default duration.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~8 minutes

Possibly related issues

• Angry-DM scorer (claude -p) hangs on combat-sprint transcripts — zero stdout for 300s+ #1040: This PR directly implements the fix described in that issue—isolating the Angry-DM lens to run sequentially after lighter lenses and increasing the timeout from 300s to 600s to prevent zero-stdout timeouts.

Possibly related PRs

• electricsheephq/WorldOS#537: Modifies the same empty-stdout/failed-attempt handling in qa/score.sh's retry loop that this PR also updates.
• electricsheephq/WorldOS#1026: Touches the same claude -p scoring invocation in qa/score.sh, adding env -u environment unsetting that this PR's changes build upon.

Poem

🐇 Three lenses to score, but one needs its space,
Angry-DM gets room, a quieter place.
The two lighter lenses run side by side,
Then Angry-DM follows with confident stride.
Six hundred seconds — enough time at last,
No more empty output, no timeout too fast! 🎯

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately captures the root cause and fix: the Angry-DM scorer timeout was too short (300s) for a legitimately slow lens (~400s), matching the PR's primary focus.
Description check	✅ Passed	The description thoroughly documents the root cause (confirmed 402s probe), the concurrency issue, the fix (600s timeout + sequential scoring), and validation steps, but does not include the CLA checklist or licensing validation section from the template.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}